5 Tips to Getting Started on the Road to AI

Data Basics, Scaling AI Christophe Balmy

This blog is a guest post from our friends at Solution BI. They are specialized in the entire business intelligence chain and the transformation of financial performance processes in the enterprise.

AI is officially everywhere —  it has played a significant role in the global health crisis both in monitoring the number of cases and tracking global lockdown and reopening patterns. A Gartner survey of "roughly 200 business and IT professionals on September 24, 2020 revealed that:

  • 24% of respondents’ organizations increased their artificial intelligence (AI) investments and 42% kept them unchanged since the onset of COVID-19.
  • Over the course of the next six to nine months, 75% of respondents will continue or start new AI initiatives as they move into the Renew phase of their organization’s post-pandemic Reset.
  • 79% of respondents said their organizations were exploring or piloting AI projects, while only 21% stated their AI initiatives were in production."*

However, some companies are still reluctant to take the plunge due to a lack of in-house expertise. With the advent of collaborative data science platforms, the barriers to AI have been greatly reduced. Read on to discover five key steps to keep in mind when starting on the AI journey.


5 Steps to Consider Before You Start

1. Take “Stock” of Your Analytics Maturity 

First, it is essential to clearly identify the added value that the AI project will bring to the company. To do this, keep yourself informed about the different use cases, meet with industry experts, or follow training courses, for example. The objective is to get a precise idea of the incremental value of an AI project on examples similar to your objective. 

It is important to start with simple projects, whose ROI is well identified beforehand. This is the only way to prove the benefits of AI. The subjects must be tangible and progress in an iterative and pragmatic way. The objective is to give confidence to the teams and to management so you can successfully gain buy-in. 

2. Don't Underestimate the Time Needed for Your AI Project

Our clients always ask themselves if it's going to take long, but, for the simplest subjects, it should only take a few weeks to get started. After a few months, we can move to a more systemized approach. Companies are increasingly aware that even with very lightweight solutions, they can start to transform themselves.

However, be careful not to minimize the duration of the project to avoid disillusionment! For example, don’t go into the project with the mindset that it will only take two weeks instead of two months because that’s unrealistic, and delays will be inevitable when that becomes reality. 


Whatever the project, AI platforms can help speed up processes to build solutions as efficiently as possible. For example, with Dataiku, a company can create a very simple system to predict its turnover for the coming months. It just needs to train the machine with its results of the previous months by uploading a simple csv file. 

Performing data transformations in Dataiku can help teams make the link between the orders from the beginning of the month and the final turnover. Everything works in drag-and-drop format and requires very little code for those who work better with data visualizations, but technical experts who prefer to code can do that — everyone can work the way they're the most comfortable.

3. Prepare Your Kickoff

While outside the scope of this post, teams need to be sure they put a process in place to calculate potential ROI from each project. Once that is done and the objective and needs from each stakeholder have been identified, the kickoff can be prepared. Very often, an AI project starts with elaboration sessions. The kickoff meetings will ensure that your AI project has enough legs to stand on — both a technical vision and a business vision. Indeed, an AI project without business sponsorship is likely to run into adoption issues. This can lead to an answer to a problem that is not a priority for the business. Often, we realize this far too late.

As mentioned before, it is interesting to start with a simple project with a reduced scope (note, however, this doesn’t mean an unreasonably short timeframe!) before leading to a Minimum Viable Product (MVP). This is a prototype on which we can then iterate, which we will grow before industrializing. 

4. The Industrialization Phase

After the use case is identified, we have demonstrated the interest of using AI tools, and have developed a small scale solution with success, we can then convince other departments (marketing, HR) to invest. Industrializing AI in a company means evangelizing it to all the company's departments and automating certain parts of the pipeline. Some examples include automating repetitive tasks like loading and processing data, running batch scoring jobs, and more. With automation in place, teams can manage more projects and scale to deliver more to production. At this point, it is the robustness of the infrastructure that is important. 

5. Avoid Technical Debt: Involve IT From the Beginning of the Project 

It is common to see projects that drag on much longer than expected. To avoid this technical debt (sustainability of IT developments over time), it is essential to involve IT very early in the project.

Business managers often consider IT as a speed bump, whereas it is essential to include it upstream of the projects. Indeed, the quality of the technical infrastructure is the condition for the success of an AI application. It is therefore essential to make an effort at the management level, so that the two talk to each other. This sometimes requires the creation of cross-functional teams.

In order to scale up, you need to have a data infrastructure that works. As soon as you want to create value, you have to do it on a robust technical foundation to integrate these functionalities into the IT system. And finally, you don't have to code everything yourself. We often use a culinary metaphor in data science: In her restaurant, a chef rarely does everything herself. For AI it's the same, at least to start with. It's even better to use off-the-shelf products and consulting partners who can show you the right roadmap before embarking on major developments.

Dataiku, Your Strongest Ally

At Solution BI, we think Dataiku has strong features to help you to start a safe AI journey:

  1. It’s easy to use — all businesses can use it because it offers a simplified vision of every AI project. Teams will have a global vision of all data projects and data journeys, from data manipulation to data visualization to the final application. 

  2. It makes scaling easy. The tool can connect, in a data-agnostic way, to any type of computational engine and any type of data (cloud, on premise, or hybrid). Snowflake is the platform most connected to Dataiku for our clients. The native Snowflake connector allows data scientists to run a query for two minutes and pay for only those two minutes of compute time. There is no reason to run a smaller cluster for four minutes anymore — spin big and spin back down when you are done!

  3. Simple and accessible, Dataiku's functionalities allow the creation of end-to-end AI projects. The tool allows the creation of final applications intended for end users, so it is adapted for the industrialization phase.

  4. The greatest quality of Dataiku is its ability to bring together collaborators in a single tool. With Dataiku, everyone collaborates in the same place, each with their own part of the project. Everyone can "see" what the others are doing, without being able to modify (if they don't have permission to). The data scientist, for example, will handle the machine learning part, the data analyst will connect to the results datasets, the sales director will have access to the dashboards, and so on.

  5. Dataiku allows the development of AI from scratch, but also enables the use of languages, libraries, plugins, and all kinds of components allowing the reuse of existing code.

*Gartner Press Release, “Gartner Survey Reveals 66% of Organizations Increased or Did Not Change AI Investments Since the Onset of Covid-19”, October 1, 2020, https://www.gartner.com/en/newsroom/press-releases/2020-10-01-gartner-survey-revels-66-percent-of-orgnizations-increased-or-did-not-change-ai-investments-since-the-onset-of-covid-19

You May Also Like

How to Reach the Apex of Data Preparation

Read More

How to Address Churn With Predictive Analytics

Read More

What Is MLOps?

Read More

4 Do's and Don'ts of Hiring and Upskilling for AI Talent

Read More